• thickertoofan@lemm.eeEnglish
    6·
    3 天前

    i have an amazing solution for this, instead of screenshotting, i save it in txt (type it out) so that later when i have a self hosted LLM assistant, i can send all the shit ive compiled till now, and ask for movies/ songs/ or any article i saved and i can just do a semantic search through it. planning to make an open source tool for this but not too good at ML

    • meme_historian@lemmy.dbzer0.comOPEnglish
      6·
      3 天前

      Oh, no need to wait for LLMs. Apache Solr should be really good at it. We used it at a company I was working at to build the most kickass search into our platform, that would actually find the stuff you were looking for…and that was back in 2018 :D

      • thickertoofan@lemm.eeEnglish
        1·
        3 天前

        ayy, that’s nice. LLMs are truely overkill just for semantic search though, didnt know there are other ways to achieve this. but we need intelligence too right. (somewhat)

        • meme_historian@lemmy.dbzer0.comOPEnglish
          3·
          3 天前

          Don’t get me wrong though… throwing an LLM at it would be a lot easier and faster. Just a mind boggling use of resources for a task that could probably be done more efficiently :D

          Setting this up with Apache Solr and a suitable search frontend runs a high risk of becoming an abandoned side project itself^^

          • thickertoofan@lemm.eeEnglish
            3·
            3 天前

            Yeah LLM seems like the go to solution. And the best one. And talking about resources, we can use barely smart models which can generate coherent sentences, be it 0.5b-3b models offloaded to CPU inference only.

        • 0xD@infosec.pubEnglish
          1·
          3 天前

          Yes, your own intelligence that you integrate into the structure of your database and queries ;)